On Tue, Apr 6, 2021 at 7:00 AM Ashish Kalra <ashish.kalra@xxxxxxx> wrote:
>
> Hello Paolo,
>
> On Tue, Apr 06, 2021 at 03:47:59PM +0200, Paolo Bonzini wrote:
> > On 06/04/21 15:26, Ashish Kalra wrote:
> > > > It's a little unintuitive to see KVM_MSR_RET_FILTERED here, since
> > > > userspace can make this happen on its own without having an entry in
> > > > this switch statement (by setting it in the msr filter bitmaps). When
> > > > using MSR filters, I would only expect to get MSR filter exits for
> > > > MSRs I specifically asked for.
> > > >
> > > > Not a huge deal, just a little unintuitive. I'm not sure other options
> > > > are much better (you could put KVM_MSR_RET_INVALID, or you could just
> > > > not have these entries in svm_{get,set}_msr).
> > > >
> > > Actually KVM_MSR_RET_FILTERED seems more logical to use, especially in
> > > comparison with KVM_MSR_RET_INVALID.
> > >
> > > Also, hooking this msr in svm_{get,set}_msr allows some in-kernel error
> > > pre-processsing before doing the pass-through to userspace.
> >
> > I agree that it should be up to userspace to set up the filter since we now
> > have that functionality.
> >
>
> The userspace is still setting up the filter and handling this MSR, it
> is only some basic error pre-processing being done in-kernel here.
The bit that is unintuitive is that userspace will still get the
kvm_exit from an msr filter for KVM_MSR_RET_FILTERED even if they did
not add it to the filters. I don't think this is a huge deal:
userspace asked for it indirectly (through cpuid+sev enablement).
>
> Thanks,
> Ashish
>
> > Let me read the whole threads for the past versions to see what the
> > objections were...
> >
> > Paolo
> >
